Software for initialized explore-exploit creates a plurality of probability distributions. Each of these probability distributions is generated by inputting a quantitative description of one or more features associated with an image into a regression model that outputs a probability distribution for a measure of engagingness for the image. Each of the images is conceptually related to the other images. The software uses the plurality of probability distributions to initialize a multi-armed bandit model that outputs a serving scheme for each of the images. Then the software serves a plurality of the images on a web page displaying search results, based at least in part on the serving scheme.
Legal claims defining the scope of protection, as filed with the USPTO.
1. A method, comprising operations of: creating a first plurality of probability distributions from a plurality of images that are distinct, wherein each of the probability distributions is generated by inputting a quantitative description of one or more features associated with one of the plurality of images into a regression model that outputs a probability distribution for a measure of engagingness for that image and wherein each of the plurality of images is conceptually related to the other images; using the first plurality of probability distributions to initialize a multi-armed bandit model that outputs a serving scheme for each of the images; serving at least one of the images on a web page displaying search results, based at least in part on the serving scheme; monitoring user feedback on the at least one served image; creating a second plurality of probability distributions based at least in part on the user feedback; updating the multi-armed bandit model with the second plurality of probability distributions; and splitting the regression model into two or more regression models, based at least in part on the user feedback for the at least one served image and image metadata associated with the at least one served image, wherein each of the operations is executed by one or more processors.
2. The method of claim 1 , wherein the search results are produced by a facted search engine.
3. The method of claim 1 , wherein the user feedback is a share to a stream associated with a profile on a social-networking website.
4. The method of claim 2 , wherein the user feedback is a mouse click.
5. The method of claim 1 , wherein the measure of engagingness is click-through rate (CTR).
6. The method of claim 1 , wherein the multi-armed bandit model employs Bayesian inference with Lagrange relaxation.
7. The method of claim 1 , wherein the regression model is trained using images of celebrities.
8. The method of claim 1 , wherein the regression model is based on gradient-boosted decision trees.
9. The method of claim 1 , wherein the quantitative representation is a feature vector.
10. A computer-readable storage medium persistently storing a program, wherein the program, when executed, instructs a processor to perform the following operations: create a first plurality of probability distributions from a plurality of images that are distinct, wherein each of the probability distributions is generated by inputting a quantitative description of one or more features associated with one of the plurality of images into a regression model that outputs a probability distribution for a measure of engagingness for that image and wherein each of the plurality of images is conceptually related to the other images; use the first plurality of probability distributions to initialize a multi-armed bandit model that outputs a serving scheme for each of the images; serve at least one of the images on a web page displaying search results, based at least in part on the serving scheme; monitor user feedback on the at least one served image; create a second plurality of probability distributions based at least in part on the user feedback; update the multi-armed bandit model with the second plurality of probability distributions; and split the regression model into two or more regression models, based at least in part on the user feedback for the at least one served image and image metadata associated with the at least one served image.
11. The computer-readable storage medium of claim 10 , wherein the search results are produced by a facted search engine.
12. The computer-readable storage medium of claim 10 , wherein the user feedback is a share to a stream associated with a profile on a social-networking website.
13. The computer-readable storage medium of claim 11 , wherein the user feedback is a mouse click.
14. The computer-readable storage medium of claim 10 , wherein the measure of engagingness is click-through rate (CTR).
15. The computer-readable storage medium of claim 10 , wherein the multi-armed bandit model employs Bayesian inference with Lagrange relaxation.
16. The computer-readable storage medium of claim 10 , wherein the regression model is trained using images of celebrities.
17. The computer-readable storage medium of claim 10 , wherein the regression model is based on gradient-boosted decision trees.
18. The computer-readable storage medium of claim 10 , wherein the quantitative representation is a feature vector.
19. A method, comprising the operations of: creating a first plurality of probability distributions from a plurality of videos that are distinct, wherein each of the probability distributions is generated by inputting a quantitative description of one or more features associated with one of the plurality of videos into a regression model that outputs a probability distribution for a measure of engagingness for that video and wherein each of the plurality of videos is conceptually related to the other videos; using the first plurality of probability distributions to initialize a multi-armed bandit model that outputs a serving scheme for each of the videos; serving at least one of the videos on a web page displaying search results, based at least in part on the serving scheme; monitoring user feedback on the at least one served video; and creating a second plurality of probability distributions based at least in part on the user feedback; updating the multi-armed bandit model with the second plurality of probability distributions; and splitting the regression model into two or more regression models, based at least in part on the user feedback for the at least one served image and image metadata associated with the at least one served image, wherein each of the operations is executed by one or more processors.
20. The method of claim 19 , wherein the search results are produced by a facted search engine.
Cooperative Patent Classification codes for this invention. Click any code to explore related patents in that topic.
March 29, 2012
December 30, 2014
Browse 5M+ US patents with plain-English claim translations and AI-generated analysis.